BIT used AI to co-create images for a campaign to encourage adults in the Middle East to share technical training opportunities with their children. Our conclusion: AI can help researchers identify hidden themes, build rapport and shed light on perspectives traditional research methods can miss.
The challenge: creating a comms campaign to promote technical training
Technical and vocational education and training (TVET) can help tackle the high youth unemployment in the Middle East by equipping young people with in-demand skills. And yet, uptake is low, in part because of negative social norms associating TVET with failure and poor awareness of course options and job opportunities available.
To address these barriers, we’re developing a comms campaign to encourage adults to share information about TVET with their kids. For it to be effective, we needed the images and messaging to resonate with our target population, and so co-creation with participants from the region was essential.
How and why we used AI
In our previous research, social desirability bias ran high. With traditional focus groups using questions about pre-pared images, we might get polite feedback that concealed much better ideas.
And so, we co-created images using software, like DALL.E, that turned participants’ prompts into images to help visualise their thoughts and enable immediate iteration based on their feedback in real-time.
How did it go?
Reading up on how to create great prompts definitely helped us generate more useful, higher quality images. But, even then we generated some horrors (Image 1).
Image 1. Prompt: A young man performing solar panel maintenance
AI added real value to our research in three ways:
1. Revealing themes through visualisation
In our first focus group, we invited a whole family to explore household dynamics. We used the AI tool to create images to accompany the promotional messages they had developed. The father first suggested “a young man in a nice modern garage working on cars”. In response to this image, the mother insisted that “the garage has to look safe, and the worker should be wearing safety gear”.
Many parents had mentioned car mechanic in our previous research as an attractive job. But, it was only when prompted with an image that we saw their worry for the child’s safety among heavy machinery. We then found this to be an important factor among other parents too.
Image 2. Prompt: Young woman wearing a head scarf working a beautiful modern beauty salon.
2. Exploring cultural sensitivities
Later, we got another family to do the same activity. Since their young daughter wanted to be a beautician, their initial prompt was “a young woman smiling working at a nice beauty salon”. When the family looked at the output (Image 2), they were confused and started shaking their heads. The father didn’t understand what was going on, and the mother explained that nail varnish invalidated the ablution before prayer.
Whilst this had not mattered in some interviews, we couldn’t risk alienating people. Further iteration identified suitable images that were clear and relatable: women blow-drying unveiled hair or putting on make-up.
3. Building rapport through laughter
AI image generation got people engaged and made them laugh. Once, we were losing participants’ attention: the son started scrolling on his phone and the dad was talking about an unrelated issue. All that changed when we introduced the AI tool. They were actively engaged in creating images, and they burst out in laughter at some of the absurd images.
The tool is not perfect, returning mangled faces, surreal object positioning and incomplete hands. But maybe that is not such a bad thing, as it brought a light and joyful atmosphere in the room.
There are plenty of challenges and risks in using AI – not least having no idea what the image will be. But, AI offers an exciting way to augment traditional qualitative research and actively involve target populations in co-creating contextually sensitive solutions.